diff options
Diffstat (limited to 'sysdeps/aarch64/memcmp.S')
| -rw-r--r-- | sysdeps/aarch64/memcmp.S | 221 |
1 files changed, 112 insertions, 109 deletions
diff --git a/sysdeps/aarch64/memcmp.S b/sysdeps/aarch64/memcmp.S index ae2d997973..743bc078bb 100644 --- a/sysdeps/aarch64/memcmp.S +++ b/sysdeps/aarch64/memcmp.S @@ -1,6 +1,6 @@ /* memcmp - compare memory - Copyright (C) 2013-2016 Free Software Foundation, Inc. + Copyright (C) 2013-2018 Free Software Foundation, Inc. This file is part of the GNU C Library. @@ -22,129 +22,132 @@ /* Assumptions: * - * ARMv8-a, AArch64 + * ARMv8-a, AArch64, unaligned accesses. */ /* Parameters and result. */ #define src1 x0 #define src2 x1 #define limit x2 -#define result x0 +#define result w0 /* Internal variables. */ #define data1 x3 #define data1w w3 -#define data2 x4 -#define data2w w4 -#define has_nul x5 -#define diff x6 -#define endloop x7 -#define tmp1 x8 -#define tmp2 x9 -#define tmp3 x10 -#define pos x11 -#define limit_wd x12 -#define mask x13 +#define data1h x4 +#define data2 x5 +#define data2w w5 +#define data2h x6 +#define tmp1 x7 +#define tmp2 x8 ENTRY_ALIGN (memcmp, 6) - cbz limit, L(ret0) - eor tmp1, src1, src2 - tst tmp1, #7 - b.ne L(misaligned8) - ands tmp1, src1, #7 - b.ne L(mutual_align) - add limit_wd, limit, #7 - lsr limit_wd, limit_wd, #3 - /* Start of performance-critical section -- one 64B cache line. */ -L(loop_aligned): - ldr data1, [src1], #8 - ldr data2, [src2], #8 -L(start_realigned): - subs limit_wd, limit_wd, #1 - eor diff, data1, data2 /* Non-zero if differences found. */ - csinv endloop, diff, xzr, ne /* Last Dword or differences. */ - cbz endloop, L(loop_aligned) - /* End of performance-critical section -- one 64B cache line. */ - - /* Not reached the limit, must have found a diff. */ - cbnz limit_wd, L(not_limit) - - /* Limit % 8 == 0 => all bytes significant. */ - ands limit, limit, #7 - b.eq L(not_limit) - - lsl limit, limit, #3 /* Bits -> bytes. */ - mov mask, #~0 -#ifdef __AARCH64EB__ - lsr mask, mask, limit -#else - lsl mask, mask, limit -#endif - bic data1, data1, mask - bic data2, data2, mask - - orr diff, diff, mask -L(not_limit): - -#ifndef __AARCH64EB__ - rev diff, diff + DELOUSE (0) + DELOUSE (1) + DELOUSE (2) + + subs limit, limit, 8 + b.lo L(less8) + + ldr data1, [src1], 8 + ldr data2, [src2], 8 + cmp data1, data2 + b.ne L(return) + + subs limit, limit, 8 + b.gt L(more16) + + ldr data1, [src1, limit] + ldr data2, [src2, limit] + b L(return) + +L(more16): + ldr data1, [src1], 8 + ldr data2, [src2], 8 + cmp data1, data2 + bne L(return) + + /* Jump directly to comparing the last 16 bytes for 32 byte (or less) + strings. */ + subs limit, limit, 16 + b.ls L(last_bytes) + + /* We overlap loads between 0-32 bytes at either side of SRC1 when we + try to align, so limit it only to strings larger than 128 bytes. */ + cmp limit, 96 + b.ls L(loop16) + + /* Align src1 and adjust src2 with bytes not yet done. */ + and tmp1, src1, 15 + add limit, limit, tmp1 + sub src1, src1, tmp1 + sub src2, src2, tmp1 + + /* Loop performing 16 bytes per iteration using aligned src1. + Limit is pre-decremented by 16 and must be larger than zero. + Exit if <= 16 bytes left to do or if the data is not equal. */ + .p2align 4 +L(loop16): + ldp data1, data1h, [src1], 16 + ldp data2, data2h, [src2], 16 + subs limit, limit, 16 + ccmp data1, data2, 0, hi + ccmp data1h, data2h, 0, eq + b.eq L(loop16) + + cmp data1, data2 + bne L(return) + mov data1, data1h + mov data2, data2h + cmp data1, data2 + bne L(return) + + /* Compare last 1-16 bytes using unaligned access. */ +L(last_bytes): + add src1, src1, limit + add src2, src2, limit + ldp data1, data1h, [src1] + ldp data2, data2h, [src2] + cmp data1, data2 + bne L(return) + mov data1, data1h + mov data2, data2h + cmp data1, data2 + + /* Compare data bytes and set return value to 0, -1 or 1. */ +L(return): +#ifndef __AARCH64EB__ rev data1, data1 rev data2, data2 #endif - /* The MS-non-zero bit of DIFF marks either the first bit - that is different, or the end of the significant data. - Shifting left now will bring the critical information into the - top bits. */ - clz pos, diff - lsl data1, data1, pos - lsl data2, data2, pos - /* But we need to zero-extend (char is unsigned) the value and then - perform a signed 32-bit subtraction. */ - lsr data1, data1, #56 - sub result, data1, data2, lsr #56 - RET - -L(mutual_align): - /* Sources are mutually aligned, but are not currently at an - alignment boundary. Round down the addresses and then mask off - the bytes that precede the start point. */ - bic src1, src1, #7 - bic src2, src2, #7 - add limit, limit, tmp1 /* Adjust the limit for the extra. */ - lsl tmp1, tmp1, #3 /* Bytes beyond alignment -> bits. */ - ldr data1, [src1], #8 - neg tmp1, tmp1 /* Bits to alignment -64. */ - ldr data2, [src2], #8 - mov tmp2, #~0 -#ifdef __AARCH64EB__ - /* Big-endian. Early bytes are at MSB. */ - lsl tmp2, tmp2, tmp1 /* Shift (tmp1 & 63). */ -#else - /* Little-endian. Early bytes are at LSB. */ - lsr tmp2, tmp2, tmp1 /* Shift (tmp1 & 63). */ -#endif - add limit_wd, limit, #7 - orr data1, data1, tmp2 - orr data2, data2, tmp2 - lsr limit_wd, limit_wd, #3 - b L(start_realigned) - -L(ret0): - mov result, #0 - RET - - .p2align 6 -L(misaligned8): - sub limit, limit, #1 -1: - /* Perhaps we can do better than this. */ - ldrb data1w, [src1], #1 - ldrb data2w, [src2], #1 - subs limit, limit, #1 - ccmp data1w, data2w, #0, cs /* NZCV = 0b0000. */ - b.eq 1b - sub result, data1, data2 - RET + cmp data1, data2 +L(ret_eq): + cset result, ne + cneg result, result, lo + ret + + .p2align 4 + /* Compare up to 8 bytes. Limit is [-8..-1]. */ +L(less8): + adds limit, limit, 4 + b.lo L(less4) + ldr data1w, [src1], 4 + ldr data2w, [src2], 4 + cmp data1w, data2w + b.ne L(return) + sub limit, limit, 4 +L(less4): + adds limit, limit, 4 + beq L(ret_eq) +L(byte_loop): + ldrb data1w, [src1], 1 + ldrb data2w, [src2], 1 + subs limit, limit, 1 + ccmp data1w, data2w, 0, ne /* NZCV = 0b0000. */ + b.eq L(byte_loop) + sub result, data1w, data2w + ret + END (memcmp) #undef bcmp weak_alias (memcmp, bcmp) |